🚀 Feature

Applying tensor.min(dim), tensor.max(dim), tensor.argmin(dim), tensor.argmax(dim) causes an error when any dimension of the tensor is of size 0.

>>> torch.zeros((2,0)).max(0)
RuntimeError: cannot perform reduction function max on tensor with no elements because the operation does not have an identity

In my opinion this error should only be thrown, if the dim is associated to a dimension which is of size 0.

Motivation

The current behavior is inconsistent with numpy and requires unnecessary case handling for empty tensors.

>>> np.zeros((2,0)).max(0)
array([], dtype=float64)

Pitch

Return empty tensors instead of raising an exception.

>>> torch.zeros((2,0)).max(0)
torch.return_types.max(
values=tensor([]),
indices=tensor([], dtype=torch.int64))